IMPACT OF CHILD DEVELOPMENT CENTERS IN THE FIRST YEAR OF PRESCHOOL1
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
ARTIGOS ARTÍCULOS ARTICLES
http://dx.doi.org/10.18222/eae.v30i73.5850
IMPACT OF CHILD DEVELOPMENT
CENTERS IN THE FIRST YEAR OF
PRESCHOOL1
MARIANE C. KOSLINSKII
TIAGO LISBOA BARTHOLOII
TRANSLATED BY: Vikas NarangIII
ABSTRACT
The paper analyzes the impact of the main program for early childhood education
adopted by Rio de Janeiro public municipal system since 2010 on children’s development
at preschool. It discusses the methodological design and results of international research
and indicates the scarcity of studies with robust designs in the Brazilian context. A
random sample of 46 schools (2.716 children) collected in two waves: at the beginning and
end of 2017, from the study “Baseline Brazil (BLB)”, was used in the statistical treatment
of data. The descriptive analysis of data suggests an association in the same direction
between attendance to the program and development in language. However, analysis
using hierarchical models including controlling variables do not indicate a statistically
significant effect of attending the program for cognitive development in the short run.
KEYWORDS LONGITUDINAL STUDY • PROGRAM EVALUATION • PRESCHOOL
EDUCATION • EARLY CHILDHOOD.
1 This paper is part of Baseline Brazil Research – a longitudinal study on the learning trajectory of children, which
was funded by the Inter-American Development Bank, the Rio de Janeiro State Research Support Fund [Fundo de
Amparo à Pesquisa do Estado do Rio de Janeiro – FAPERJ], the Alfa and Beto Institute and the National Council for
Scientific and Technological Development [Conselho Nacional de Desenvolvimento Científico e Tecnológico – CNPq]
I Federal University of Rio de Janeiro (UFRJ), Rio de Janeiro-RJ, Brazil; https://orcid.org/0000-0002-9644-5041;
mckoslinski@gmail.com
II Federal University of Rio de Janeiro (UFRJ), Rio de Janeiro-RJ, Brazil; http://orcid.org/0000-0002-2400-8707;
tiagobartholo@gmail.com
III Editage, Brazil; contato@editage.com
280 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932XIMPACTO DOS ESPAÇOS DE DESENVOLVIMENTO INFANTIL
NO PRIMEIRO ANO NA PRÉ-ESCOLA
RESUMO
O artigo analisa o impacto do principal programa para a educação infantil, adotado pela
rede municipal da cidade do Rio de Janeiro a partir de 2010, no desenvolvimento cognitivo
das crianças da pré-escola. Discute o desenho metodológico, resultados de pesquisas
internacionais e a escassez de estudos com desenhos robustos no contexto brasileiro. Uma
amostra aleatória de 46 escolas (2.716 crianças), em duas ondas: início e final de 2017,
pertencente ao estudo Linha de Base Brasil (LBB), foi empregada no tratamento estatístico
de dados. A análise descritiva dos dados sugere uma associação de mesmo sentido entre
a frequência ao programa e o desenvolvimento em linguagem. No entanto, análises
multivariadas utilizando modelos hierárquicos com diversos controles não indicam um
efeito do programa no desenvolvimento cognitivo no curto prazo.
PALAVRAS-CHAVE ESTUDO LONGITUDINAL • AVALIAÇÃO DE PROGRAMAS •
EDUCAÇÃO PRÉ-ESCOLAR • PRIMEIRA INFÂNCIA.
IMPACTO DE LOS ESPACIOS DE DESARROLLO
INFANTIL EN EL PRIMER AÑO PREESCOLAR
RESUMEN
El artículo analiza el impacto del principal programa para la educación infantil adoptado
por la red municipal de la ciudad de Río de Janeiro desde 2010 para el desarrollo cognitivo
de los niños preescolares. Discute el diseño metodológico, los resultados de investigaciones
internacionales y la escasez de estudios con diseños robustos en el contexto brasileño. Una
muestra aleatoria de 46 escuelas (2.716 niños) en dos olas: inicio y fines de 2017, que
pertenece al estudio Linha de Base Brasil (LBB), se utilizó en el tratamiento estadístico
de datos. El análisis descriptivo de dichos datos sugiere una asociación de mismo sentido
entre la participación en el programa y el desarrollo en lenguaje. Sin embargo, análisis con
múltiples variaciones que utilizan modelos jerárquicos con diversos controles no indican
un efecto del programa en el desarrollo cognitivo a corto plazo.
PALABRAS CLAVE ESTUDIO LONGITUDINAL • EVALUACIÓN DE PROGRAMAS •
EDUCACIÓN PREESCOLAR • PRIMERA INFANCIA.
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 281INTRODUCTION
Several studies conducted in different contexts have indicated that preschool
attendance is an effective way of ensuring decrease of inequality in educational
opportunities. It contributes to various dimensions of child development, as
well as longer school trajectories, especially for vulnerable children from
families with low socioeconomic status. International longitudinal studies
have confirmed that, as a general rule, children who have had the opportunity
to attend good quality early childhood-care programs have shown greater
development of cognitive and socioemotional skills in the short and medium
term, during their school trajectories (PEISNER-FEINBERG et al., 2001; SYLVA et
al., 2010; SAMMONS et al., 2006; TYMMS et al., 2009, NICHD, 2006).
An important discussion centers around the definition of quality of care
in early childhood education. The international literature describes two
main educational factors associated with the development of children at the
beginning of schooling: A) structure quality, which includes teacher training,
attendance, the adult-child ratio, and building characteristics; and B) quality
of school processes, a category that includes adult-child interactions, the
instructional climate, and pedagogical practices. Studies with more robust
designs, which include controls for previous skills and the sociodemographic
282 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932Xcharacteristics of families and children, suggest that the quality of school
processes is moderately or weakly associated with the cognitive, linguistic,
and socioemotional development of children, both during preschool and in
later education. However, the evidence that relates to the impact of the quality
of the care structure is not as consistent (HOWES et al., 2008; SYLVA et al., 2006,
2010; NATIONAL INSTITUTE OF CHILD HEALTH AND HUMAN DEVELOPMENT
EARLY CHILD CARE RESEARCH NETWORK – NICHD ECCRN; DUNCAN, 2003;
WALSTON; WEST, 2004).
Early childhood education in Brazil has achieved considerable progress
nationally, since it was inserted into the basic education system by the 1988
Constitution and consolidated in the 1996 Law on the Guidelines and Bases
of Education. The expansion of early childhood education demonstrates
the increasing importance given to educating young children. The growing
enrollment and rate of attendance is evident from time series. Only 13.8%
of Brazilian children under the age of 3 years attended preschool in 2001;
this figure reached 30.4% in 2015. In 2001, 64.4% of 4–5 year-olds attended
preschool; this figure reached 90.5% in 2015 (OBSERVATÓRIO DO PNE, s.d.).
However, these percentages are still far short of Goal 1 of the National
Education Plan, which aims to have universal pre-schooling by 2020, in
accordance with Law No. 12.796, which makes it compulsory for children to
attend school from the age of four (BRASIL, 2013) and expects 50% of children
under the age of 3 years to have the option to attend crèche. The municipal
systems responsible for early childhood education have encountered many
challenges in guaranteeing a greater number of places. Frequently, unplanned
expansions have led to low-quality services, including poorly maintained
institutions, which are detrimental to children’s development (CAMPOS et
al., 2011a). However, few studies have investigated the impact of preschool/
early childhood education and/or quality of care on children’s development.
The few studies that do exist are not longitudinal; for this reason, they have
limitations when it comes to drawing causal relationships between preschool
quality and the future learning trajectory of children.
The present study investigates the impact of the main program for early
childhood education adopted in the city of Rio de Janeiro – Child Development
Center [Espaço de Desenvolvimento Indantil – EDI] – on the development of
children in the first year of preschool. The study is restricted to a sample of
schools in the municipal system of Rio de Janeiro and is limited to observing
the impact of distinct forms of care within this system. The program
guidelines are primarily associated with the structure of care, in that they
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 283include “a combination of crèche and preschool, primary care classrooms,2
infant libraries, and full-time care” (RIO DE JANEIRO, 2010). The first section
of this paper discusses the designs and results of international studies that
have focused on the relationship between the quality of the preschool care
structure and/or process on children’s development and subsequent education.
It also reveals the lack of studies with designs robust enough to identify the
impact of educational policies for early childhood education in the Brazilian
context. The second section presents the results and the analyses carried out
to observe the impact of EDIs on the cognitive development of children.
Specifically, it analyzes the data collected in the research project Baseline
Brazil [Linha de Base Brasil- LBB]: a longitudinal study on children’s learning
trajectory, which used a probabilistic sample representative of the municipal
public system of Rio de Janeiro, with 46 schools (2,716 children), stratified by
type of care (regular or EDI). Data were collected in two waves during 2017
– in March 2017 (at the start of the school year) and in November/December
2017 (at the end of the school year). A descriptive data analysis indicates a
non-uniform implementation of the program. Some EDIs operate part-
time and/or in old buildings – without the infrastructure and classroom
time considered adequate by the program guidelines. A descriptive analysis
also suggests an association in the same direction between attending EDIs
and language development in children. However, analyses of the results
using hierarchical linear models, with controls related to the children’s
characteristics, their families, the schools and, especially, the measurement
of each child’s initial level of development upon entering preschool (Wave
1) do not indicate an effect of attending EDI schools on children’s cognitive
development, at least not in the short term. The study has detected variations
in program implementation; however, the results observed are consistent
even when the analyses take into account only the effect of attending EDIs in
full-time classrooms (the child-and-class model).
2 The primary care classrooms include arrangements for “the presence of health workers who can provide first-aid for
day-to-day situations; register and follow-up on children’s growth/development; organize health records and referrals
when an initial diagnosis is made” (RIO DE JANEIRO, 2010).
284 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932XTHE EFFECTS OF SCHOOL AND QUALITY OF CARE ON EARLY CHILDHOOD
EDUCATION: THE EVIDENCE AND DEBATE
For Goldstein (1997), in a “scientifically ideal” world, the study of causality in
education would be based on randomized experiments in which individuals
(students, teachers, and principals) and “treatments” (school factors, such as
classroom size, curricular content, pedagogical material, school organization,
and school composition) would be randomly distributed to institutions. This
would make it possible to guarantee study designs with high internal validity,
avoiding selection bias (CANO, 2009; SHADISH; COOK; CAMPBELL, et al., 2001;
MURMANE; WILLET, 2011).3
In the real world, however, it is common to have no control over the
distribution of individuals, the composition of schools and classes, or the
way in which teachers teach. When it comes to theme of the present study,
there is scant robust experimental evidence on the relationship between the
characteristics of preschool quality of care and the development of children.
In other words, few studies have attempted to experimentally manipulate
the child-adult ratio and/or the educational level and training of teachers or
caregivers.4 Most of the studies that have observed the impact of quality of care
on children’s education or of school characteristics on children’s development
and/or other school outcomes are correlational. As a result, there are always
limitations or concerns that either selection bias or omitted variables in the
estimated models could explain the observed relationships (GOLDSTEIN, 1997;
NICHD ECCRN; DUNCAN, 2003; LEE, 2004; FRANCO; BROOKE; ALVES, 2008).
For Goldstein (1997), school-effectiveness studies need satisfactory
designs/models to observe the relationship between educational factors
and particular outcomes/results. To avoid selection bias and make “fair
comparisons” between schools, it is necessary to control the characteristics
that are known to have an impact on results, when using available theories.
For example, a wide range of studies have shown that students input (gender,
background, and ethnicity, for example) and initial ability/performance differ
between schools, due to various factors. To avoid selection bias, it is necessary
to measure these dimensions accurately at the individual level (FITZ-GIBBON,
1996; GOLDSTEIN, 1997; LEE, 2004). Thus, the literature on school effectiveness
3 Other models capable of improving the quality of causal inference discussed by the authors include, for example,
regression discontinuity designs and the use of instrumental variables.
4 Some examples are the evaluation of the Carolina Abecedarian Project (CAMPBELL et al., 2001) and the High Scope
Pre-school Intervention (SCHWEINHART; WEIKART, 1990).
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 285proposes some minimal conditions to draw satisfactory conclusions about the
impact of school factors on learning and other educational outcomes. The first
condition relates to the research design. It is important for the design to be
longitudinal, taking two or more measurements from the same individuals
over time.5 This is a fundamental way to control for preexisting differences
between students in cross-sectional studies, as they would not be able to account
for or make adjustments to student entrance characteristics, for example
(FITZ-GIBBON6, 1996; GOLDSTEIN, 1997). The second condition requires the use
of a multilevel regression model to investigate the “differential effectiveness”
of schools and/or teachers. The model must take into account the fact that
students are grouped in certain schools and classrooms and are, therefore,
subject to the specific influences of that school or classroom/group. Goldstein
(1997) has even suggested that replication occurs in time and space7 and the
researcher should present a plausible theory to explain the results obtained.
Vendell’s bibliographic review (2004) of the impact of different forms
and quality of early childhood care on the development and educational
trajectory of children, indicates how recent studies have approached these
minimal conditions. The first correlational studies, carried out in the 1980s,
avoided selection bias by controlling some demographic characteristics of
children and their families. More recent studies have incorporated a larger
number of covariates; some have minimized the risk associated to omitted or
unmeasured variables by using measures of the children’s prior performance
as controls or models to estimate gains (delta).
NICHD ECCRN and Duncan (2003) conducted a methodological exercise
to illustrate the limits of correlational studies, especially the risk of selection
bias. Their research used data from the National Institute of Child Health
and Human Development (NICHD)’s longitudinal “Study of Early Child
Care,”8 conducted in the United States. The objective of the exercise carried
5 Lee (2004) points out some limitations of longitudinal studies: increased cost of data collection and sample
mortality. The latter is an important limit since children with greater mobility between schools generally have a lower
socioeconomic status and/or a greater probability of belonging to racial or ethnic minorities.
6 In addition, cross-sectional data refer to a learning aggregate over time; however, measures of school conditions
(such as those collected by the Basic Education Assessment System – Saeb – and Prova Brasil) refer to the year in
which the data were collected, and therefore a temporal synchronization between the measures would be lacking
(FRANCO; BROOKE; ALVES, 2008).
7 Replication is defended as studies of school effectiveness observe large fluctuations of school value-added
measures, calculated from residual models, over time (GOLDSTEIN, 1997; GORARD, 2011).
8 The study recruited a sample of 1,354 children and followed them from birth until 54 months of age. Extensive data
were collected on the children at different stages (15, 24, 36, and 54 months of age), including data on their family
and environment; quality, quantity, and type of care received; and cognitive, language, and socio-behavioral skills.
286 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932Xout by NICHD ECCRN and Duncan (2003) was to observe the impact of
school-attendance quality9 in early childhood on the cognitive and language
performance of children at 24 and 54 months of age, using several models. The
findings showed that a lack of adequate controls could generate exaggerated
estimates or overestimated parameters.
The regression models for estimating children’s cognitive and language
scores at 54 months initially used only variables related to the type, quality,
and quantity of early childhood care. Over time, they gradually added
covariates related to the children’s characteristics and their families. The
results show a decrease in the estimated coefficients for the quality of care,
as controls were inserted in relation to the children’s characteristics and
family environments. However, the more extensive family controls added to
later models, do not seem to alter the quality-of-care coefficients obtained.
The same estimation exercise was carried out in various models, using as
a control variable the children’s results at 24 months and, later, using as
a dependent variable the children’s gains between 24 and 54 months. In
models that controlled for previous performance and family characteristics,
the coefficients decreased even more, in comparison to models without this
control. In the models that showed gains (delta) between 24 and 54 months,
the insertion of family controls decreased the estimated quality-of-care
coefficients to the point where they were no longer statistically significant
(NICHD ECCRN; DUNCAN, 2003).
More recent studies, with longitudinal designs, large samples, low
sample mortality, controls for the characteristics of children and families,
and/or multilevel models, have focused on the impact of the type of care
(public, private, or offering other types of service), the quality of the structure
(e.g., teacher training, care period, adult-child ratio) and classroom processes
(including interactions, the instructional climate, and pedagogical practices10)
on the development of children in various dimensions.
In measuring the quality of the care process, the results of several studies
seem to converge. For example, analyses of the results using data collected
9 To measure the quality of care, the study used the Observational Record of Caregiving Environment (ORCE).
10 In these studies, process quality is measured using classroom observation instruments, such as the Early Childhood
Environment Rating Scale (ECERS) – which includes sub-scales on space and furniture, personal care routines,
language-reasoning, learning, interaction, program structure, and parents and staff – and the Classroom Assessment
Scoring System (CLASS) encompassing dimensions such as emotional support, classroom organization, and
instructional support (HARMS; CLIFFORD; CRYER, 2005; PIANTA; PARO; HANRE, 2008).
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 287by the Effective Pre-School and Primary Education (EPPE)11 project in the United
Kingdom observed that the quality of preschool care has a small (effect sizes
ranging from 0.11 to 0.20) but statistically significant impact12 on the cognitive
and social-behavioral progress of children (SYLVA et al., 2010; SYLVA et al.,
2006). The models used pre-test controls (performance or scores of children’s
ability at preschool entrance) and the children’s and families’ characteristics,
and home learning environment. In addition, the study observed lasting
impacts, mainly from attending a high-quality preschool, on children’s results
in primary education (SAMMONS, 2006; SYLVA et al., 2010).
Similarly, Peisner-Feinberg et al. (2001) observed a small association in
the same direction (effect sizes of 0.03 to 0.18), between the quality of the
preschool care process13 and the development of children’s language, reading,
and mathematics skills, as well as14 a moderate association in the same
direction for sociability and attention. Studies conducted within the NICHD
project found that the quality of early childhood care was modestly associated
(effect size between 0.10 and 0.11) with children’s academic and language
outcomes at 54 months (NICHD, 2006). Both studies used several controls
related to characteristics of children and their families. Howes et al. (2008)
observed an association between children’s development of language, literacy,
math, and socio-behavioral skills in the short term, during preschool (pre-
kindergarten programs) and the quality of classroom processes and school
structures.15 The study estimated hierarchical models, which used not only
a pre-test control variable, but also the children’s short term cognitive gains
as a dependent variable. Morover, the analyses considered many controls
related to children and their families and the results obtained revealed gains
in language, reading and social behavior, especially in programs with higher
process quality. The authors observed similar results in analyses of gains that
11 The EPPE followed a random sample of 2,800 children enrolled in various establishments in England from preschool
entrance (children from 3 years to 4 years and 3 months) to the end of primary school stages 1 and 2 (7 and 11 years,
respectively) (SYLVA et al., 2010).
12 The quality of the preschool care process was measured using two instruments: the Early Childhood Environment
Rating Scale – Revised Edition (ECERS-R) and the Early Childhood Rating Scale – Extension (ECER-E).
13 The quality of the process was measured from the Early Childhood Environment Scale (ECERS), Child Caregiver
Interaction Scale (CCIS), Early Childhood Observation Form and Adult Involvement Scale (AIS) instruments.
14 This is part of the Cost Quality, and Child Outcomes in Child Care Centers Study, conducted in four US states. The
results indicate the impact of classroom practices and teacher-child relationships on the development of children’s
language, cognitive and social skills.
15 The process quality measures were obtained from the Early Childhood Environment Rating Scale-Revised (ECERS-R),
Classroom Assessment Scoring System (CLASS), and Emerging Academic Snapshot.
288 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932Xused a pre-test as a covariate16; however, the coefficients/parameters obtained
in the gain (delta) models brought smaller or more conservative estimates
(effect sizes ranging from 0.06 to 0.13) for the effects of preschool quality
on language and reading than models that used previous performance as a
covariate (effect sizes ranging from 0.06 to 0.19). The same trend was observed
when using models that estimated progress in mathematics and children’s
social-behavior skills.
If the results of studies with robust designs that examined the relationship
between the quality of the early childhood-care process (adult-child interaction,
instructional climate, and pedagogical practices) – measured using various
observational instruments – and cognitive development in children’s language,
mathematical skills, and social behavior seem to converge, the same trend
cannot be observed in studies that focus on the quality of the care structure
(e.g., teacher training, attendance time, the adult-child ratio, and building
characteristics). For example, in assessing quantity of care, the results of the
studies carried out by NICHD (2006) revealed a non-expected impact on
the behavior of children who attended preschool for longer hours. In the UK,
the EPPE project did not observe an impact of full-time attendance (SYLVA
et al., 2010); the Early Childhood Longitudinal Study, carried out in the
U.S. with kindergarten children (5-year-olds), observed greater progress in
language and mathematics among children who attended full-time programs
than those who attended part-time programs (WALSTON; WEST, 2004).
In the area of teacher training, the EPPE project observed that preschool
care delivered by more qualified staff had higher process quality; in such
establishments, the children made more progress (SYLVA et al., 2010). Howes
et al. (2008) observed positive and statistically significant correlations between
teacher qualifications, the period of care and children’s language and reading
development. However, when covariates related to children’s characteristics
and process quality were introduced, variables related to the quality of the
care structure lost statistical significance.
In Brazil, the discussion about evaluating and setting goals for early
childhood education has revolved around access and the quality of the care
structure. The National Education Plan provides crèche and preschool access
targets, as previously mentioned. The first proposal produced by the National
16 Both models estimated dependent variables at the student level.
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 289System of Evaluation of Basic Education17 (Sinaeb) to carry out a National
Assessment of Infant Education (Anei) focused on conditions such as “physical
infrastructure, staffing, management, pedagogical resources, and accessibility,
among other relevant contextual indicators” (BRASIL, 2016). The proposal for
evaluating early childhood education in the new Basic Education Assessment
System (Saeb) is still under development. However, preliminary information
indicates that the evaluation will not include results derived from cognitive
tests; data will be collected through questionnaires distributed to teachers,
principals, and leaders (BRASIL, 2018). Considering the limitations of data
collection through questionnaires (FRANCO et al., 2003), this assessment of
early childhood education is unlikely to adequately or accurately capture or
measure classroom and/or child development processes. Even with the recent
expansion of access and discussions about the parameters for evaluating early
childhood education, the impact of quality of care on children’s development
has not been studied in a systematic way.
A few studies have focused on the possible effects of crèche and/or
preschool on educational results and trajectories. For example, a study by
Damiani et al. (2011) observed that children who attended preschool tended
to have longer educational trajectories. Campos et al. (2011b) investigated the
relationship (in three Brazilian capitals) between the quality of the preschool
process (measured using the Ecers-R scale) on outcomes for Portuguese
language learners in the second year of elementary school (measured by
Provinha Brasil). The data indicated that children who attended institutions
with good process quality performed better on the test than children who
did not attend preschool or who did so in institutions with unsatisfactory
quality levels (CAMPOS et al., 2011b). However, the study did not have a
longitudinal design; it used a cross-classified hierarchical model to estimate
the performance of children assessed by Provinha Brasil, without controlling
for family characteristics, elementary schools attended, and most importantly,
the children’s previous development. The analyses had limitations to avoid
selection bias. In addition, the internal validity was greatly affected by the
large sample attrition.
17 Sinaeb was a proposal to expand the Brazilian education evaluation systems in accordance with the conceptions and
goals of the National Education Plan. For more details of the proposal, consult Santos, Horta Neto and Junqueira
(2017).
290 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932XSTUDY DESIGN AND DATA COLLECTION
In the present study, we were initially interested in investigating the impact
of a recent program aimed at early childhood education, called “Child
Development Center” (EDI). It has been implemented gradually since 2010 by
the Rio de Janeiro Municipal Department of Education. The program created
a difference in the quality of the care structure for early childhood education,
since the EDIs had the following characteristics, which distinguished them
from regular early childhood-care schools: “a combination of crèche and
preschool, primary care spaces, infant libraries and full-time care” (RIO DE
JANEIRO, 2010). New schools were built with a specific architecture and
adequate spaces to serve this stage of basic education.
The principal objective of the present study is to observe the impact of
attending EDIs (as opposed to regular schools) on the cognitive development
of children in the first year of preschool. We emphasize that, as the study
sample includes only municipal public schools and regularly enrolled
students, it can only verify the impact of attending different types of care in
the public system. It cannot compare care in public and private schools or
observe the effects on those who did or did not attend preschool.18 To address
the study question, we estimated several hierarchical regression models,
which gradually incorporated variables related to attending EDIs, children’s
characteristics, their family background, and measurements of their cognitive
development at the beginning of preschool.
The data were collected by the study “Baseline Brazil” (Linha de Base Brasil
– LBB), a large-scale longitudinal study on the effects of preschool and the first
year of elementary school, which aims to follow children from 4 to 7 years of age.
The study selected a random probabilistic sample of 46 schools (approximately
2,700 children) from the municipal system of Rio de Janeiro, stratified by type
of service offered and Regional Education Coordinating Unit.19 Specifically, in
the municipal public system of Rio de Janeiro, the study considered two types of
schools: A) EDIs – established in accordance with the principal public policy for
early childhood education in the city; B) regular schools offering early childhood
education only or preschool and elementary school classes.
18 In future stages of the study, we intend to include a sample of private schools, which will allow a greater comparison
of the effects of different types of preschool offer. Furthermore, we intend to use regression discontinuity models to
differentiate the preschool effect from the maturation effect. This type of analysis has already been performed by
Tymms, Merrell and Henderson (1997) using data collected in England.
19 In the municipal public network of Rio de Janeiro, there are 11 Regional Education Units with a very diverse number of
schools and enrolled children.
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 291Figure 1 shows the different stages of data collection. The two waves of
collection in March and November/December 2017 were carried out when the
children were 4/5 years of age, at the beginning and end of the school year.
FIGURE 1 – DESIGN OF THE LONGITUDINAL STUDY
1st year
Preschool I Preschool I Preschool II
Elementary
O1 O2 O3 O4
4 years 4/5 years 5/6 years 6/7 years
2017 2019
Source: Compiled by the authors.
In each wave, we collected data on the children’s cognitive development
and their fine and gross motor skills. The cognitive data were collected
using Performance Indicators in Primary Schools (Pips), an accurate tool that has
been tested for 20 years in the UK and other countries. It creates20 a baseline
measure at the beginning of the compulsory schooling process to monitor
the children’s development.21 Pips consists of the following dimensions:
A) Writing; B) Vocabulary; C) Ideas about reading – evaluating concepts about
print; C) Phonological awareness; D) Letter identification; E) Words recognition
and reading; F) Ideas about mathematics; G) Counting and numbers;
H) Addition and subtraction without symbols; I) Identification of forms; and
J) Identification of numbers. Language and mathematics scores were estimated
using items on the Pips cognitive test, via the Rasch model (Boone, 2006; Bond
and Fox, 2015) and Winsteps software. The present study removed the items
that involved reading phrases at the time of calculation because only a very
few children (around 1.0% of the sample) in the first year of pre-school are
exposed to this section of the test.
The test was applied individually; the duration varied between 10 and 20
minutes. In Brazil, the test was applied using booklet (with images and text
20 Pips has already been adapted by several countries especially those part of the iPips (Placing Early Childhood
Education at the Heart of Worldwide Policy Making) study. The coordinators of the iPips study, at the University of
Durham, work together with teams from the participating countries in the process of adaptation and translation of
Pips. For a more detailed description of the adaptation procedures of Pips employed by the Rio de Janeiro study, see
Bartholo et al. (forthcoming).
21 Pips is not an instrument that measures curricular contents but contains dimensions with high predictive capacity of
cognitive development during the schooling process.
292 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932Xpositioned in front of the child); and a test application version that runs on a
tablet and guides the researcher on items to be presented. The application is
adjusted in accordance with the child’s responses. The procedure continues
when consecutive, correct answers are given; it is interrupted when the child
demonstrates a lack of knowledge of particular content. Each test session
presents increasingly difficult items, allowing the test to have the minimum
desired duration, without leaving the child bored by many simple or overly
difficult questions (TYMMS; MERREL; JONES, 2004).
Table 1 shows the initial number of sampled children in Wave 1 of the
study, mobility during the first year of the longitudinal study, and attrition
between Waves 1 and 2 (at the beginning and end of 2017):
TABLE 1 – Description of number of cases in Wave 1 and Wave 2 and attrition
NUMBER PERCENTAGE
OF CASES
SME children in the 1st wave 2,716 100.00%
SME children in the 2nd wave 2,848 -
Not found in the system 106 3.9%
Migrated to other Municipal Department of Education (SME) schools 164 6.0%
Students in the sample schools in the 2nd wave 2,446 90.1 %
Total children measured twice 2,237 82.4%
Source: Compiled by the authors.
Approximately 4% of the children were missing from the Academic
Management System (Sistema de Gestão Acadêmico – SGA). Our team had
access to this system which allowed us to observe the entire trajectory of
children within the municipal public schools. The missing children had
moved out of the city or migrated to the private schools. These were difficult
situations because there was no information in the SGA on their new schools.
However, our most substantial loss involved children who migrated to other
schools (which were not sample schools) in the Municipal Department of
Education of the Municipality of Rio de Janeiro (SME-RJ) and children who
remained in the same schools but missed classes on a recurring basis. We
observed that 90.1% of the children who participated in the 1st wave were still
enrolled in the schools of the study sample. However, only 82.4% of the same
total participated in the 2nd wave. Our team made multiple visits to the same
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 293schools and still had difficulty finding some children. Student absenteeism
was not an initial focus of the study, but it caught the attention of researchers
and required adjustments to the data collection strategies. The attrition
reported in the study (17.5%) is considered low to moderate.
In addition to the Pips cognitive tool, the study also collected contextual
data from the children (a questionnaire completed by parents/guardians)22 and
schools. We have collected contextual information using the questionnaires
from approximately 62% of the guardians of children in the sample. To
minimize the loss of cases, missing data were complemented with data
provided by the SGA of the Municipal Department of Education. Therefore, the
estimated hierarchical models only used variables present in both databases
to minimize the amount of missing data and to maximize the total number
of children.
When we began the descriptive analyses of the distribution of classes
and children among sample schools, we observed a variation in the program
implementation, given the guidelines provided by the Municipal Department
of Education. For example, while not all EDI classrooms were full-time, some
regular schools (not EDIs) also offered full-time provision. Likewise, not all
EDIs operated in new buildings, built specifically to serve early childhood
education.
TABLE 2 – Distribution of schools and classrooms according to the type of service
offered (EDI and non-EDI); care period (full-time and part-time)
TOTAL FULL-TIME NEW BUILDING
EDI NON EDI EDI NON EDI EDI NON EDI
School 19 27 - - 10 2
Classroom 73 43 20 13 42 3
Children measured twice 1,309 928 394 263 716 63
Source: Compiled by the authors, based on data from SGA and LBB research.
This variation led us to carry out other analyses, not just observing the
impact of attending an EDI school, but also observing the impact of attending
an EDI that strictly followed the program guidelines in the program (full time
and/or a new building). It is interesting to note that, although there are only
22 Children’s contextual data were collected from a questionnaire for parents/guardians applied by researchers at the
time of entry and exit from school and on parent/teacher day meetings.
294 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932X19 EDI schools, these schools offer more pre-school classrooms than regular
schools. Finally, we observed a differentiated service provision within the
same EDI: full-time and part-time classes.
ANALYSIS, RESULTS, AND DISCUSSION
Table 3 describes the variables and the data sources used in the analyses; Table 4
presents descriptive statistics based on those variables.
TABLE 3 – Description of the variables used in the analyses
NAME TYPE DESCRIPTION SOURCE
DEPENDENT VARIABLE
Lang2 Continuous Measure of language in Wave 2 LBB Project
EXPLANATORY VARIABLES LEVEL 2 – SCHOOL/CLASSROOM
Indicates whether the school is an EDI or a regular
EDI Dummy SME
network school
Indicates whether the classroom is in an EDI is full-
Full-time EDI Dummy SME
time (1 = full-time EDI, 0 = other classrooms)
Proportion of children with at least one parent/
Proportion of high-school
Continuous guardian with a high-school and/or higher level SGA/SME and LBB
educated parents
education
Proportion non-white Continuous Proportion of non-white children SME
Proportion of children whose parents are
Proportion in poverty Continuous SGA/SME and LBB
beneficiaries of cash transfer programs
EXPLANATORY VARIABLES – LEVEL 1 – CHILD
Lang1 Continuous Measure of language in Wave 1 LBB
Indicates if child has special educational needs
Special needs Dummy LBB
(0 = no; 1 = yes)
Gender Dummy Indicates the gender of the child (0 = girl; 1 = boy) SGA/SME
Indicates whether parents/guardians of the child
Poverty Dummy are beneficiaries of cash transfer programs (0 = no; SGA/SME and LBB
1 = yes)
Indicates whether at least one parent/guardian of
High-school Educated
Dummy the child has completed high school and/or higher SGA/SME and LBB
Parents
level education (0 = no; 1 = yes)
Indicates the color of the child (declared by
Non-white Dummy SGA/SME
parents/guardians) (0 = white; 1 = not white)
Age Continuous Indicates the child’s age in months SGA/SME
Source: Compiled by the authors.
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 295TABLE 4 – Descriptive statistics of the variables used in the analyses
STANDARD
VARIABLE N MEAN
DEVIATION
CHILDREN VARIABLES
Lang2 2,237 0.25 1.00
Lang1 2,237 -0.35 0.89
Special needs 2,237 0.02 -
Gender 2,237 0.52 -
Poverty 2,237 0.31 -
High-school Educated Parents 1,809 0.57 -
Non-white 2,125 0.63 -
Age 2,226 54.28 4.03
SCHOOL/CLASS VARIABLES
EDI (school) 46 0.41 -
Proportion with high-school education 46 0.45 0.15
Proportion non-white (school) 46 0.59 0.14
Proportion in poverty (school) 46 0.31 0.14
Full-time EDI (class) 116 0.17 -
Proportion High-school Educated
116 0.43 0.19
Parents (class)
Proportion non-white (class) 116 0.60 0.13
Proportion poverty (class) 116 0.25 0.17
Source: Compiled by the authors based on data from SGA and LBB research.
Table 5 reveals a small difference between the profiles of children
attending EDIs and those attending regular schools. We observed that a
higher proportion of children in EDIs (60%), had parents who had completed
high school, than did those who attended regular schools (53%); there was a
higher proportion of non-white children in regular schools (66%) than in EDIs
(62%). The language starting point of children at EDIs was also slightly higher.
A similar pattern, slightly more pronounced, was observed when children
who attended EDIs full-time were compared to children who attended other
schools/classes in the sample.
296 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932XTABLE 5 – Mean of the child-level variables according to type of provision and care period
NOT FULL-
EDI NON-EDI FULL TIME EDI
VARIABLE TIME EDI
(SCHOOL) (SCHOOL) (CLASSROOM)
(CLASSROOM)
CHILDREN’S VARIABLES (MEAN)
Lang2 0.30 0.20 0.34 0.21
Lang1 -0.33 -0.38 -0.25 -0.38
Special needs 0.02 0.02 0.03 0.02
Gender 0.51 0.52 0.51 0.53
Poverty 0.31 0.31 0.27 0.35
High-school Educated Parents 0.60 0.53 0.62 0.55
Non-white 0.62 0.66 0.56 0.65
Age 54.17 54.38 53.9 54.4
Source: Compiled by the authors based on data from SGA and LBB research.
One of the aims of this study was to understand the pattern of attrition.
Table 6 presents the language and mathematical measures for children who
participated only in Wave 1, as well as for those who participated in the two
waves of data collection. The mean and standard deviations were calculated
using Wave 1 measurements to observe whether the group that participated
in two waves was similar to the group that only participated in Wave 1:
TABLE 6 – Wave-1 Measurements: children who participated only in Wave 1 and
children measured twice23
LANGUAGE WAVE 1* MATHEMATICS WAVE 1
STANDARD STANDARD
MEAN MEAN
DEVIATION DEVIATION
Children who participated only in Wave 1 -0.53 1.00 -1.76 1.37
Children measured twice (Waves 1 and 2) -0.39 1.01 -1.54 1.36
Source: Compiled by the authors.
*The T-test for equality of means indicates that the language-related difference in means between the two
groups was statistically significant (p < 0.01).
The data suggest that there is a small, statistically significant, difference
between the two groups, with a pattern indicating that the group that
participated in both waves had a higher mean starting point. One possible
23 The language and mathematics measures used in the table were calculated using Rasch model and considering only
Wave 1 results.
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 297explanation is that the present study lost children who were no longer in
school and who were therefore less likely to participate in the two research
waves. The initial descriptive analysis did not allow us to rule out the risk of
bias in the study’s attrition.
Another source of attrition in the study was the difficulty encountered
when trying to find parents or guardians at the school. Many of the children
were privately driven to school or went with an older sibling. We were able to
distribute the questionnaire to 62% of the guardians and to use the SGA data to
complete part of the missing information on key family characteristics. Table 7
presents the mean and standard deviation for Wave 1 data, comparing children
with and without information on their parents’ education. A preliminary
analysis of the data suggests that loss had a pattern. Children who lacked
information about their parents’ education had a lower starting point (Wave
1). The data suggest that the loss of information about family characteristics
was probably not random; it could affect the estimates of hierarchical models.
However, the difference in the observed means was not statistically significant.
TABLE 7 – Language and mathematical measures for wave 1 for children with and
without information on their parents’ education
LANGUAGE WAVE 1* MATHEMATICS WAVE 1*
STANDARD STANDARD
MEAN MEAN
DEVIATION DEVIATION
Children with information on parents’ education -0.40 1.00 -1.55 1.35
Children with no information on parents’ education -0.48 1.00 -1.67 1.40
Source: Compiled by the authors.
*T test for equality of means indicates that the difference between the two groups is not statistically significant
(p > 0.05).
We, initially, performed a descriptive analyses of the progress in language
and mathematics made by children attending pre-school in EDIs and in regular
schools within the municipal public system. We also carried out descriptive
analyses, considering the variations observed in the implementation of the
program (EDIs offered both full-time and part-time classes; some were housed
in new buildings and others in old buildings, adapted to receive the program
or not). We then, in order to make inferential analysis, we used hierarchical
linear regression models estimating the development of children in Wave 2.
We gradually inserted controls related to the children’s attendance at EDIs, the
composition of schools/classrooms, the children’s characteristics, their family
background, and cognitive measurements from Wave 1 (entry into pre-school).
298 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932XGRAPHS 1 and 2 – Mean mathematics and language measures for waves 1 and 2 for EDIs
and regular schools
Mathematics wave 1 e wave 2 Language wave 1 and wave 2
-0,5 0,4
1 2 0,2
-1
0
-1,5 1 2
-0,2
-2 -0,4
-2,5 -0,6
EDI Non EDI EDI Non EDI
Source: Compiled by the authors.
GRAPHS 3 and 4 – Mean mathematics and language measures for waves 1 and 2 for
children in EDIs, new and buildings
Mathematics wave 1 and wave 2 Language wave 1 and wave 2
0 0,4
1 2 0,2
-1 0
1 2
-0,2
-2
-0,4
-3 -0,6
Old building New building Old building New building
Source: Compiled by the authors.
GRAPHS 5 and 6 – Mean mathematics and language measures for waves 1 and 2 for
children who attended full-time classrooms in EDIs and in other classrooms
Mathematics wave 1 and wave 2 Language wave 1 and wave 2
0 0,5
1 2
-1
0
-2 1 2
-3 -0,5
Full-time EDI Other classrooms Full-time EDI Other classrooms
Source: Compiled by the authors.
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 299As noted in the previous graphs, children attending EDIs appeared to
have made slightly greater progress in language than those attending regular
pre-schools (Graph 2). However, progress in math seems similar, with a small
advantage for children who attended regular schools (Graph 1). Graphs 3
and 4 also present analyses by school and suggest that children attending
pre-schools in new buildings had, on average, slightly reduced development
in language and mathematics, in comparison to their peers who attended
municipal pre-schools in old buildings.
Graphs 5 and 6 present analyses at a class level of the services offered
to children attending a full-time class at an EDI, in comparison to other
offerings: part-time EDI, full-time at a regular pre-school, and part-time at a
regular pre-school. The idea was to see whether there was an effect associated
with attending pre-school that met the key policy requirements.
From the trends observed in the graphs, two-level linear hierarchical
regression models were used to estimate only the language measure for
children in Wave 2 (November/December 2017). The aim was to verify whether
the impact of the EDIs was observed even after controls were included that
related to pupil composition, the children’s characteristics and families, and
children’s development in language Wave 1. The first level refers to children
and the second level to schools.
Five models were estimated using the following specifications:
Model 1: LING2ti = π0i + eti (level 1 – child)
π0i = β00 + β01*(EDIi) + r0i (level 2 – child)
In the first model, we only inserted the variable indicating whether the
school was an EDI or a regular school.
Model 2: LING2ti = π0i + eti
π0i = β00 + β01*(EDIi) + β02*(Prop High-school Educated Parentsi)
+ β03*(Prop non-whitei) + β04*(Prop non-whitei) + r0i
In the second model, we added, in Level 2 (school) variables related to
the proportion of children whose parents had completed secondary or higher
education; the proportion of non-white children; and the proportion of
children in a situation of poverty (part of a cash-transfer program).
300 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932XModel 3: LING2ti = π0i + π1i*(Special Needsti) + π2i*(Genderti) + π3i*(Povertyti)
+ π4i*(High-school Educated Parentsi) + π5i*(Non-whiteti) + π6i*(Ageti) + eti
π0i = β00 + β01*(EDIi) + β02*(Prop High-school Educated Parentsi) +
β03*(Prop non-whitei) + β04*(Prop Poverti) + r0i
π1i = β10
π2i = β20
π3i = β30
π4i = β40
π5i = β50
π6i = β60
Model 3 added variables related to the children (special needs, gender,
color, and age) and their families (whether parents had completed high school
and/or higher education; whether they were part of a cash-transfer program).
Model 4 added Level 1 controls related to the child’s level of development
in Wave 1 (Lang 1); Model 5 was equivalent to Model 4, however, Level 1 had
no controls for the child’s color or parents’ education. This leaner model
was estimated when insertion of two variables implied a loss of 428 children
(approximately 20% of the sample).
Table 8 presents the results of parameters obtained from models with
different effect sizes. Effect sizes are simple ways to report differences
between two groups, rather than just discussing statistical significance. There
is an extensive debate on how to interpret effect sizes (Hattie, 2009; Higgins,
Kokotsaky, and Coe, 2012). We chose to use the classification present in
Higgins, Kokotsaky, and Coe (2012): (i) effect sizes up to 0.18 are considered
small; (ii) from 0.19 to 0.44 moderate; (iii) from 0.45 to 0.62 high; and
(iv) greater than 0.70 are very high.24 The effect sizes were calculated according
to the methodology developed by Tymms (2004) and Tymms, Merrell, and
Handerson (1997).25
24 The authors suggest an interpretation of effect sizes in months of educational progress, considering an effect size of
a standard deviation as equivalent to one year of instruction in elementary education.
25 According to the methodology suggested by the authors, the continuous independent variables were standardized,
and the coefficients reported in effect sizes of these variables express the differences estimated for individuals with
one standard deviation above and one standard deviation below the mean.
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 301TABLE 8 – Linear hierarchical regression models estimating language measures in Wave 2
(effect sizes)
MODEL 1 MODEL 2 MODEL 3 MODEL 4 MODEL 5
SCHOOL (Level 2)
EDI 0.126* 0.071 0.095 0.113 0.097
Proportion of high-school
0.150* -0.027 0.006 0.123
educated parents
Proportion non-white 0.069 0.074 0.078 0.083
Proportion in poverty -0.209** -0.213** -0.297*** -0.272***
CHILD (Level 1)
Language – Wave 1 1.833*** 1.909***
Special needs -1.140 ***
-0.737 ***
-0.532***
Gender -0.123*** -0.011 -0.034
Poverty -0.111 **
-0.037 -0.054
High-school Educated
0.372*** 0.189***
Parents
Non-white -0.064 -0.024
Age 0.672 ***
0.297*** 0.240***
Explained variation
Level 1 0 0 14.6 49.5 50.0
Level 2 4.4 19.5 16.9 51.2 59.4
Null model
r0 0.041
E 0.754
ICC 5.2
Source: Compiled by the authors.
Note: * p < 0.10, ** p < 0.05, *** p 0.01.
In Table 8, the estimated coefficients for children frequenting EDIs
are positive (effect size of 0.113 in model 4).26 However, the values are not
statistically significant for the 46 schools sampled in the study.
The first point to note is that only 5.2% of the variance is explained at Level 2
of the model, which suggests that, at the end of the first year of preschool,
most observed differences could be explained at Level 1 of the model (child).
These results will be explored in more depth in another study that uses a
regression discontinuity design to analyze the overall impact of attending
preschool in the municipal public system.
26 According to the interpretation of Higgins, Kokotsaky and Coe (2012), this effect would be small and roughly
equivalent to two months of progress.
302 Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-6831, e-ISSN 1984-932XModel 3 presents results (commonly reported in educational research in
Brazil and abroad) used to identify factors associated with child development.
There is, however, an important limitation in the model, namely: the absence
of an initial measure of the development of the child. The comparison
of the coefficients of Models 3 and 4 highlights exactly the importance of
using longitudinal designs to estimate the effect of educational policies and
programs. The introduction of the baseline (Wave 1 language) significantly
alters both the explanatory power of the model (the proportion of variance
explained in Level 1 increases by 339%) and the coefficients of the other
covariates. All lost explanatory power and some items, such as gender and
poverty, cease to be statistically significant.
Also noteworthy is the great impact of age on the model. Even controlling
for the baseline measure (Wave 1), age has an effect size of 0.297. This result
can be explained by the so-called maturation effect, which tends to be larger in
young children. This result has been little debated or explored by academics,
principals, pedagogical coordinators, and teachers who work in the schools.
Our study suggests that the child’s birth month is a key variable for estimating
the starting and ending point of children cognitive development at the first
year of preschool. This information should be taken into account by early
childhood education teachers and teachers in the early years of elementary
school, when making decisions about children who need special attention
and about a child’s failure (specifically in elementary school).
The next table also present the results of two-level hierarchal linear
regressions. In this case, however, the first level refers to children and the
second to classrooms. This second set of analyses allows us to discern a possible
policy effect by identifying children in full-time EDI classrooms (format
stipulated in the SME-RJ documents). This model has another advantage,
which is the number of analysis units (classrooms) at the second level – a total
of 116.
Estud. Aval. Educ., São Paulo, v. 30, n. 73, p. 280-311, jan./abr. 2019, ISSN 0103-683, e-ISSN 1984-932X 303You can also read
